Fabio Valentini wrote:
This sounds like you didn't read my whole post. Because I don't want to make updates going from "pending -> stable" directly impossible,
Well, several of the people who replied do want that, if I understand their replies correctly.
I think your original proposal that only blocks the automated push, requiring a manual push, would be workable. (I think autokarma pushes are a bad idea to begin with.)
Still, it is more than what would be required to address the technical issue at hand.
I just want it not to happen unless necessary, because it drastically reduces the exposure of updates to broader user testing.
It is always a tradeoff between getting urgent fixes out fast and waiting for more users to test them.
That isn't that big of a problem most of the time, since "fedora" / "updates" and "updates-testing" repositories are composed daily, but during freezes, this leads to the weird problem that possibly important updates get stuck in a state where they are available from *no repository at all*.
That is the real problem that needs fixing, and the fix for that would be for Bodhi to:
- if an update is in pending → stable state for Fedora n, AND
- if Fedora n is currently in a freeze (and ONLY in that case), THEN
- push the update to testing instead AND
- keep it queued for stable, i.e., put it into testing → stable state.
But there is no valid reason to do that for releases that are not frozen and where the update can just go out directly to stable.
Why should updates for "stable" releases receive less testing coverage than those for a pre-release? That doesn't make any sense.
Because the pre-releases are in a freeze that prevents getting the updates out to stable and the stable releases are not. My argument is a purely technical one, not a policy one.
Kevin Kofler